What ValidDraft records while you write - and what it never does

It records that a key was pressed - not which one

While you write, ValidDraft keeps a typing-evidence journal: a stream of timing events - a keystroke, a pause, a correction. For ordinary letters, the journal records that a key was pressed, not which one. Editing keys such as Backspace, Enter, and the arrows are named in the record, because corrections are part of what makes writing look human. Rhythm is captured; content is not.

What the analysis reads is the shape of composing - pace, pauses, bursts, corrections - not the composition. That is the whole trick: the evidence that a person wrote it lives in how the text arrived, so the record never needs to quote what the text says.

Nothing is recorded until you say yes

Behavioral capture starts behind a consent screen titled “Before you start writing” - unchecked by default, with a plain “Not now” to decline. The screen states the boundary in the product’s own words: the data is “never used for identification, advertising, or tracking outside this purpose,” and “the raw keystroke log itself is deleted immediately after scoring.” You can withdraw consent at any time from your profile - withdrawing blocks new writing sessions, and it does not affect certificates you’ve already earned.

The privacy policy carries the formal version, including its “We do not” list - no advertising use, no selling data, no training general-purpose AI on your content. And verification involves no camera of any kind - that has its own post.

Your words stay home until you seal

In ValidDraft for Mac, manuscript chapters live on your Mac’s disk, and the app is built for offline writing. Sealing is the one moment your words leave your machine: the chapter text and its journal go to the server to be analyzed, because the scoring runs there. Until you click Seal, going offline pauses nothing but uploads - the evidence trail keeps building locally.

Even the writing helper respects the boundary. Writing Assist runs “100% on-device. Nothing ever leaves your computer” - and every accepted suggestion is noted in your writing record, so the assistance itself is disclosed rather than hidden. One honest caveat: a sealed chapter’s text is stored alongside its certificate so the certificate stays verifiable. Stored is not shown - which is the next section.

What the world sees when you publish

The publish dialog spells it out before you commit. “What becomes public (anyone with the link):” “Your manuscript title, your name, and which chapters were verified.” And then the sentence this post exists to explain: “Your actual writing is never shown on the certificate” - “proving a human wrote it without revealing a word of it.”

The certificate’s process panel shows the session’s rhythm and shape only. Your numeric humanity score is yours alone: scores never appear on public pages - the public claim is deliberately binary, verified human, with no number for a stranger to misread. The public checker follows the same rule. It is a yes-or-no certificate lookup - in its own words, “Not a plagiarism tool. Not an AI detector.”

You keep the off switches

Publishing is reversible. Unpublish at any time and “the public certificate goes offline immediately” - your sealed chapters are kept, and re-publishing later mints a fresh certificate with a new link while “the old link stays dead.” You can delete a verification, and you can request a complete JSON export of your account data - documents, analyses, verifications, consents - from settings. The certificate proves a human wrote it, and it does that job without revealing a word of it.